Opto-Electronic Engineering, Volume. 52, Issue 7, 250055(2025)
A few-shot optical fingerprint liveness detection method diagram of the overall framework
Figures & Tables(16)
Table 1. Comparison of three transforms in fingerprint image processing
View tableView in ArticleTable 1. Comparison of three transforms in fingerprint image processing
Item Fourier transform Discrete cosine transform Wavelet transform Representation form Global frequency Global frequency Multi-scale (local + global) Output type Complex Real Complex or real (depends on implementation) Energy concentration Moderate High Moderate (depends on wavelet bases) Spatial localization Poor Poor Strong Computational complexity Relatively high Relatively low High Targeted feature properties Global frequency analysis Image compression, detail enhancement Multi-resolution feature extraction Typical applications Frequency domain analysis Local texture discrimination Multi-scale feature modeling
Table 2. Dataset sample distribution
View tableView in ArticleTable 2. Dataset sample distribution
Dataset Sensor Train set Test set Live Spoof Live Spoof LivDet 2013 Biometrika 1000 1000 1000 1000 CrossMatch 1250 1000 1250 1000 Italdata 1000 1000 1000 1000 Swipe 1220 979 1153 1000 Total 4470 3979 4403 4000 LivDet 2015 Biometrika 1000 1000 1000 1500 CrossMatch 1510 1473 1000 1500 GreenBit 4440 4495 1000 1500 Total 6950 6968 3000 4500
Table 3. Values of hyperparameters
View tableView in ArticleTable 3. Values of hyperparameters
Hyperparameter Value Batch size 32 Epochs 200 Learning rate 0.001 Number of support samples K 5, 10 Feature matching parameter M 4 Feature matching parameter L 2 Feature matching parameter C 2 Number of layers in MLP module 3 Number of nodes in MLP module 64
Table 4. Detection results on the LivDet 2013 dataset (K=5)
View tableView in ArticleTable 4. Detection results on the LivDet 2013 dataset (K=5)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 0.50 0.40 0.45 99.55 CrossMatch 0.40 0.40 0.40 99.60 Italdata 0.30 0.20 0.25 99.75 Swipe 0.20 0.43 0.32 99.67 Overall 0.35 0.36 0.36 99.64
Table 5. Detection results on the LivDet 2013 dataset (K=10)
View tableView in ArticleTable 5. Detection results on the LivDet 2013 dataset (K=10)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 0.30 0.30 0.30 99.70 CrossMatch 0.20 0.32 0.26 99.73 Italdata 0.20 0.10 0.15 99.85 Swipe 0.10 0.17 0.14 99.86 Overall 0.20 0.23 0.21 99.79
Table 6. Detection results on the LivDet 2015 dataset (K=5)
View tableView in ArticleTable 6. Detection results on the LivDet 2015 dataset (K=5)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 0.47 1.10 0.78 99.28 CrossMatch 0.80 0.70 0.75 99.24 GreenBit 0.60 0.60 0.60 99.40 Overall 0.62 0.80 0.71 99.31
Table 7. Detection results on the LivDet 2015 dataset (K=10)
View tableView in ArticleTable 7. Detection results on the LivDet 2015 dataset (K=10)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 0.33 0.60 0.47 99.56 CrossMatch 0.40 0.50 0.45 99.56 GreenBit 0.27 0.60 0.43 99.60 Overall 0.33 0.57 0.45 99.57
Table 8. Test comparison results between the proposed detection method and related works
View tableView in ArticleTable 8. Test comparison results between the proposed detection method and related works
Method Dataset Few-shot ACER/% DeFraudNet (2020) [ 27 ]LivDet 2013 No 0.28 DeFraudNet (2020) [ 27 ]LivDet 2015 No 0.84 DRBM + DBM (2020) [ 28 ]LivDet 2013 No 3.60 ALDRN (2020) [ 29 ]LivDet 2013 No 2.96 LivDet 2015 No 5.32 Method based on Fisher vector (2021) [ 30 ]LivDet 2013 No 0.30 LivDet 2015 No 1.46 MFFFLD (2022) [ 31 ]LivDet 2013 No 0.96 LivDet 2015 No 0.91 HyFiPAD (2022) [ 32 ]LivDet 2013 No 2.88 LivDet 2015 No 2.97 ViT Unified (2023) [ 33 ]LivDet 2013 No 1.91 LivDet 2015 No 0.48 LFLDNet (2023) [ 34 ]LivDet 2015 No 2.20 LFLDNet + CycleGAN (2023) [ 34 ]LivDet 2015 No 1.72 Method based on Res-CNN (2024) [ 35 ]LivDet 2015 No 2.13 MoSFPAD (2025) [ 36 ]LivDet 2013 No 0.22 LivDet 2015 No 2.45 Proposed method (K=5) LivDet 2013 Yes 0.36 Proposed method (K=5) LivDet 2015 Yes 0.71 Proposed method (K=10) LivDet 2013 Yes 0.21 Proposed method (K=10) LivDet 2015 Yes 0.45
Table 9. Cross-sensor detection results on the LivDet 2013 dataset (K=10)
View tableView in ArticleTable 9. Cross-sensor detection results on the LivDet 2013 dataset (K=10)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 2.30 1.20 1.75 98.25 CrossMatch 1.40 1.44 1.42 98.58 Italdata 2.30 3.20 2.75 97.25 Swipe 4.40 5.12 4.76 95.22 Overall 2.60 2.75 2.67 97.32
Table 10. Cross-sensor detection results on the LivDet 2015 dataset (K=10)
View tableView in ArticleTable 10. Cross-sensor detection results on the LivDet 2015 dataset (K=10)
Sensor APCER/% BPCER/% ACER/% ACC/% Biometrika 5.13 11.40 8.27 92.36 CrossMatch 4.60 9.50 7.05 93.44 GreenBit 4.80 8.60 6.70 93.68 Overall 4.84 9.83 7.34 93.16
Tools
Get Citation
Copy Citation Text
Congyuan Xu, Jun Yang, Panpan Li, Kun Deng. A few-shot optical fingerprint liveness detection method diagram of the overall framework[J]. Opto-Electronic Engineering, 2025, 52(7): 250055
Paper Information
Category: Article
Received: Feb. 26, 2025
Accepted: Apr. 25, 2025
Published Online: Sep. 4, 2025
The Author Email: Congyuan Xu (许聪源)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20